Lesson 5 of 5
Quality controls
9 min
Listen to this lesson
Integration and data architecture is only as good as its controls. Data enters through many doors, travels through interfaces and lands in analytics. Controls at each stage stop bad data early, detect what slips through and prove that what arrived equals what was sent. Architects specify these controls; testers and data owners prove they work.
Controls along the data path
- EntryValidation and derivation rules, mandatory fields, duplicate checks, approval for sensitive changes
- TransferSchema validation, mapping checks, guaranteed delivery, error queues with owners
- ArrivalReceiver confirmations, application-level checks, rejection with a reason
- ReconciliationCounts, amounts and key totals compared between sender and receiver
- MonitoringDashboards and alerts on failures, backlogs and data quality KPIs
Preventive controls at entry
The cheapest defect is the one never created. In S/4HANA, master data governance can apply validation and derivation rules, duplicate checks and approval workflows before a record is activated. For transactional interfaces, the receiving application's own checks apply: an inbound sales order is validated like one entered by Maria's team. SAP Master Data Integration's documentation also lists schema validation among its features. Specify, per flow, which checks run and what happens to rejected data.
Detective controls in transit
Every asynchronous interface needs monitoring with an owner. SAP Cloud ALM's Integration and Exception Monitoring links individual messages into end-to-end flows across cloud services and systems, lets users search for a message by business context such as an order number (if the component sends that context with the message) and raises alerts for error conditions. It maps each component's statuses into common status groups such as ERROR, WARNING, IN PROCESS and SUCCESSFUL. Design the business context attributes into each interface from the start; adding them later is much harder.
Reconciliation
Reconciliation proves completeness and accuracy by comparing sender and receiver. It works at three levels: counts (records sent equals records received), totals (sums of amounts or quantities agree) and keys (the same set of identifiers on both sides). Reconciliation is required for migration loads and valuable for daily interfaces with financial impact.
| Flow | Count check | Total check | Key check |
|---|---|---|---|
| Wave 2 customer load | 1,946 customers in scope = 1,946 business partners created | Not applicable | Every in-scope legacy number appears as a reference |
| Wave 2 open AR load | Open items per customer | MXN 48,615,320.40 in legacy = open AR in company code 1910 at the key date | Legacy invoice numbers carried as references |
| MES production confirmations | Confirmations per shift | Quantity confirmed per order | Order numbers |
| Bank statement files | Statements per account | Opening + movements = closing balance | Statement numbers in sequence |
Contoso reconciliation controls by flow.
Data quality KPIs
- Completeness: share of records with all mandatory attributes for their use.
- Uniqueness: suspected duplicates per thousand business partners.
- Timeliness: age of messages waiting in error queues; replication lag to the data platform.
- Accuracy: share of sampled records that match their source documents.
- Ownership: open data quality issues per data owner, with age.
What would you do?
Two weeks after go-live, finance finds that 37 MES production confirmations never reached S/4HANA. They had failed with a mapping error and sat in an error queue without alerts. What is the most important architectural fix?
- Specify controls at entry, transfer, arrival, reconciliation and monitoring.
- Prevent at entry with validation, duplicate checks and approvals.
- Monitor every async flow with an owner, alerts and business context for search.
- Reconcile counts, totals and keys; track data quality KPIs per owner.
Check your understanding · not graded
Lesson check
1. What three levels does reconciliation compare?
2. What does SAP Cloud ALM Integration and Exception Monitoring need to find a message by order number?
3. What is the target figure for reconciling the Wave 2 open AR load?
4. Which is a preventive control?